Composites: Part A 90 (2016) 22–32

Contents lists available at ScienceDirect

Composites: Part A
journal homepage: www.elsevier.com/locate/compositesa

A thermo-viscoelastic approach for the characterization and modeling

of the bending behavior of thermoplastic composites
Steffen Ropers a,⇑, Marton Kardos a, Tim A. Osswald b
a
Volkswagen AG, Group Research, Postbox 14990, 38440 Wolfsburg, Germany
b
Polymer Engineering Center, University of Wisconsin-Madison, 1513 University Avenue, Madison, WI 53706, USA

a r t i c l e i n f o a b s t r a c t

Article history: Currently, there is no established and cost-effective method for the bending characterization of continu-
Received 19 January 2016 ous fiber reinforced thermoplastic composites. Isothermal mechanical testing techniques are time and
Received in revised form 6 April 2016 labor-intensive and deliver information only about distinct points of the temperature-dependent
Accepted 18 June 2016
property curves. In this study, Dynamic Mechanical Analysis (DMA) as well as novel rheometer-based
Available online 24 June 2016
bending experiments were performed to assess temperature-dependent and viscoelastic behavior. On
the basis of the experimental results a new method was defined and validated for the efficient
Keywords:
characterization of temperature-dependent elastic bending behavior via DMA. Furthermore a linear
A. Thermoplastic resin
B. Thermomechanical
viscoelastic material model was derived from DMA experiments by means of time–temperature
C. Finite element analysis (FEA) superposition. As the material behavior proved to be of a highly viscoelastic nature, a method was
E. Forming developed to calibrate a material model, the parallel rheological framework, implemented in Abaqus.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction subversions is the most commonly used manufacturing technique

The automotive industry is endeavoring to reduce the curb
weights of the vehicles using lower density materials and
employing lightweight design techniques to meet the stringent
safety requirements while reducing emissions. A great number of
non- and semi-structural components have already been replaced
by lighter materials, mostly with light metals and unreinforced or
fiber-reinforced plastics. Continuous fiber-reinforced plastics exhi-
bit excellent mechanical properties and are capable of substituting
metals in structural applications. The aerospace industry has been
using these materials for decades, but the automotive segment is
significantly more cost-sensitive [1]. Furthermore, international
legislations require new vehicles to be recyclable to a given weight
percentage. Thermoset resin systems do not possess a re-melting
capability and are therefore difficult to recycle. Thermoplastic
composites solve this problem and simultaneously offer relatively
fast cycle times compared to thermoset systems, making them
very interesting for the automotive industry. Since a trial and error
based development is insufficient and costly, virtual prototyping is
vital to enable their industrial implementation and cost-effective
development. The thermoforming process along with several
⇑ Corresponding author.
E-mail addresses: steffen.ropers@volkswagen.de (S. Ropers), marton.kardos@-
volkswagen.de (M. Kardos), tosswald@wisc.edu (T.A. Osswald).
to form continuous fiber-reinforced thermoplastic sheets into com-
plex geometries. The thermoforming process is governed by the
temperature change of the blank; the organic sheet is first heated
above the matrix’ melting temperature, then placed into the mold
and formed with relatively low forces. While and after forming, the
sheet rapidly loses heat due to radiation, convection and primarily
heat transfer to the tempered molds. Eventually it consolidates and
the part is ready for demolding. The changing temperature alters
the mechanical properties during the forming process. Thus, as
pointed out by preceding studies, thermal and mechanical analyses
need to be fully coupled in the forming simulation of a thermo-
forming process [2,3].
As shown in several studies [2,4–6], the outcome of numerical
simulations depends significantly on the mechanical properties
taken into account, hence, properties need to be thoroughly char-
acterized, along with their temperature and other dependencies.
Boisse et al. [4] studied the separated influence of tensile stiff-
nesses, in-plane shear and out-of-plane bending stiffnesses on
the forming of a simple geometry and proved the importance of
all three properties. The draping over double curved geometries
requires in-plane shear deformations [2,4] and as the shear stiff-
ness increases with increasing shear angle, wrinkles start to form.
The onset and development of wrinkle formation is a global phe-
nomenon that depends on all of the stiffnesses of the material,
however the size and shape of wrinkles is predominantly

http://dx.doi.org/10.1016/j.compositesa.2016.06.016
1359-835X/Ó 2016 Elsevier Ltd. All rights reserved.
 S. Ropers et al. / Composites: Part A 90 (2016) 22–32 23

determined by the out-of-plane bending stiffness [2,4,5,7–9]. Sev- the University of Twente, which is based on the Kawabata Evalua-
eral studies [9,10] have demonstrated the importance of bending tion System and delivers promising results [18]. This latter method
stiffness in the simulation of a forming process. Since wrinkling is capable of investigating the bending stiffness at distinct temper-
is one of the most frequently occurring failures during draping, atures and deformation velocities. The rheometer bending setup
the problem needs to be addressed. was also employed in the framework of this study to acquire the
As pointed out, that material properties need to be understood bending stiffness at predefined temperatures as control values, as
and characterized in order to be able to conduct simulations with discussed in detail in Section 6.
reasonable outcomes. The simulation scale and the material model
determine the involved list of parameters. In a previous study [6]
3. DMA material characterization
the simulation software PAM-FORM, developed by the ESI Group,
was investigated with respect to the influence of different input
Dynamic Mechanical Analysis (DMA) is a mechanical testing
parameters, such as mechanical and thermal material properties.
technique that applies an oscillating deformation to a sample
The simulation is capable of the thermomechanical modeling of
under predefined environmental conditions and analyzes the
the thermoforming process. The Design of Experiments method
material response [19], to study the elastic and viscoelastic proper-
was employed to identify the most influential parameters whose
ties of the material. The method allows for the investigation of a
characterization is worth investing resources into. The Young’s
variety of mechanical and thermal properties and their tempera-
modulus, shear modulus, bending stiffness, specific heat, tool-ply
ture and frequency dependency. DMA instruments are becoming
frictional coefficient and the heat transfer coefficient were proven
even more versatile as state-of-the-art machines offer quasistatic
to be heavily influential on the simulation outcome.
measurement modes to obtain Young’s and other moduli [20],
Internal research at Volkswagen [11] as well as Melo and Rad-
along with creep and relaxation modes [21].
ford [12] have shown that some properties, such as the Young’s
DMA measures the elastic behavior of the material, i.e. stiffness,
modulus in fiber direction, are fiber dominated, meaning that they
and describes it with the storage modulus E0 , which is the in-phase
exhibit only a small temperature dependency in the investigated
component of the reaction to the applied sinusoidal deformation.
temperature range. It is sufficient to only characterize these prop-
The loss modulus E00 is the out-of-phase component describing
erties at distinct temperatures with conventional mechanical test-
the material’s ability to dissipate energy into heat, and the ratio
ing methods. Other properties, however, are matrix dominated and
of the loss to the storage modulus is the so-called tand, i.e. damping
therefore heavily temperature-dependent. Such properties are
[19].
transverse Young’s modulus, shear modulus and out-of-plane
Researchers use DMA to characterize polymers usually below
bending stiffness. The characterization of the shear modulus and
their melting temperature. Melo et al. [12] investigated the vis-
the bending stiffness is particularly challenging, especially due to
coelastic properties of unidirectional carbon fiber reinforced PEEK
their additional deflection and shear rate dependencies. There is
tapes, but only at a temperature range below the matrix materials
no established method for the efficient characterization of these
glass transition temperature (T g ). Other studies [22,23] employed
properties [8].
DMA to examine the temperature-dependent elastic moduli of
epoxies and to compare them to conventional mechanical testing
data. They found discrepancies between the actual values but good
2. Methods for bending characterization
correlation between their temperature dependencies. Margossian
et al. [5] presented a DMA-based characterization method for the
The bending stiffness of continuous materials is classically
bending behavior of unidirectional composite tapes. The study
derived from their in-plane properties [13]. However, the investi-
was conducted under the assumption that experimental forming
gations of Yu et al. [14] have shown that the calculated bending
methods are mostly modeled as isothermal processes. Thus, the
stiffness is unrealistically high compared to the experimentally
study focuses on characterizing the material’s bending behavior
obtained value of a fiber-reinforcement. This can be attributed to
at distinct temperatures, instead of doing temperature sweeps over
sliding mechanisms between the yarns and individual fibers of
a broader range. Such a characterization method yields insufficient
the reinforcement, which assign a specific bending behavior to
data for thermomechanical simulations. A great number of isother-
the reinforcement that is independent of the tensile and shear
mal measurements would be required to characterize the temper-
properties and, therefore, cannot be deduced from the calculated
ature range in question, which brings us back to the inefficiency of
values [9,15]. Modern simulation software and material models
common mechanical testing techniques.
are aware of that and require an independent bending stiffness
The method proposed in this study enables the fast and efficient
to be defined for main fiber directions. Thus, bending characteriza-
characterization of both temperature and frequency dependencies
tion is a vital requirement for forming simulations.
of the bending stiffness, however, it does not deliver the actual
Only few studies have been carried out to date on the bending
bending stiffness values. The technique requires the bending stiff-
characterization of continuous fiber reinforced thermoplastic com-
ness to be determined at least at one temperature within the inves-
posites in their molten state [5,9,10,16]. Conventional bending test
tigated range.
methods are inapplicable to materials with low bending stiffness
such as dry reinforcement fabrics and molten state thermoplastic
prepregs. The testing techniques developed by the textile industry 4. Materials and instruments
were used as the basis for the development of techniques applica-
ble to such materials. Most bending characterization techniques All experiments were performed on two materials; both
are based on the cantilever or the Kawabata tests, which both orig- representing the most commonly used thermoplastic prepreg
inate from the textile industry. Several studies employed a modi- types. The first investigated material is a woven textile reinforced
fied version of the cantilever test to characterize the bending thermoplastic prepreg (TPP) material, Tepex 102 dynalite-RG600
behavior, such as those conducted by Boisse [4] or Peirce [17]. Mar- (x)/47% supplied by Bond-Laminates. The prepreg consist of a 2/2
gossian [5] presented an extensive review of the existing test twill glass fiber weave impregnated with a polyamide 6 (PA6)
methods for bending characterization, explaining various versions matrix. Main properties of the material are listed in the material
of the cantilever test, the Kawabata test and the V-bending test. data sheet provided by the manufacturer [24] and listed in Table 1.
The study also mentions the modified rheometer developed by The second material is a continuous unidirectional carbon fiber
24 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
Table 1
Main material properties.
Trade name Tepex dynalite Ultratape
102-RG600(x)/47% B3EC12 UD02
Matrix PA6 PA6
Fibers Glass Carbon
Reinforcement type Twill weave 2/2 UD
Fiber volume (%) 47 50
Density (g/cm3) 1.8 1.46
Thickness (mm) 0.5 0.16
reinforced thermoplastic composite tape with a polyamide 6
matrix, called Ultratape B3EC12 UD01 black supplied by BASF.
Basic material properties can be found in its material data sheet
and are also listed in Table 1.
Since the resulting storage modulus values are only used for
comparison and not to determine the actual bending stiffness,
the specimens were not conditioned to control their moisture con-
tent. However, prior to testing each specimen was stored under the
same conditions to ensure that the results are comparable.
The DMA instrument used to conduct the experiments was a
Netzsch DMA 242 Artemis E. The dynamic deformation amplitude
of the machine is limited to 240 lm, and the maximum dynamic
force is 12 N. Several deformation methods and corresponding
sample holders are available for the device. One of the basic out-
puts of the analysis is the storage modulus, which is conceptually
similar to other moduli. Depending on the deformation method
used, the storage modulus may represent tensile, shear or flexural
modulus [19,22]. The other two basic outputs are loss modulus and
damping.
Mechanical bending experiments were conducted on a modi-
fied rheometer to characterize the bending behavior of both inves-
tigated materials. The device was developed at the University of
Twente by Sachs [18] based on the Kawabata Evaluation System
(KES). The rheometer is equipped with a thermal chamber, which
enables a precise temperature control throughout the experiment.
A special fixture was designed for the rheometer; see Fig. 1. As the
rotating part of the fixture moves on a circular path, the free arc
length changes with the angle a. The specimen is not clamped
but resting loosely in the fixture on support pins to avoid tensional
or compressional strains in the specimen. The device allows for
testing at various temperatures while different rotational velocities
can be applied to deform the specimen in order to identify strain
rate dependencies of the bending behavior. Sachs used this device
to test a thermoplastic UD tape with different thicknesses (number
of plies) and at different rotational velocities and temperatures.
5. Implementation
Standard DMA cantilever-bending tests were conducted, where
minuscule sinusoidal forces, or deflections, are applied to the sam-
Fig. 1. Experimental set-up in rheometer for bending experiments [18].
ple while continuously sweeping through a temperature range. The
rheometer bending experiments were conducted in a standard
rheometer instrument employing a custom built bending fixture
that allows for the measurement of the material’s response to sub-
stantial bending deformation under isothermal conditions.
5.1. DMA experiments
Three of the available test methods are applicable to the charac-
terization of out-of-plane bending: three-point bending, single-
cantilever and dual-cantilever. Most previous studies employing
DMA used the three-point bending deformation method
[5,12,22], motivated by the elimination of clamping effects inher-
ent in the cantilever clamping system. However, when conducting
dynamic measurements, three-point-bending is not applicable to
the investigated materials above melting temperature. The sample
is not clamped by the three-point-bending fixture, only supported.
Therefore, the sample will not follow the oscillating stamp once
the sample loses most of its elastic bending properties due to being
heated above melting temperature, creating a gap between sample
and stamp and corrupting the test. Then again, cantilever clamping
systems introduce an additional shearing component to the distor-
tion [5,19,12,22]. This issue can be overcome by introducing shear
correction to the calculation. However, this study uses DMA results
for comparison, thus introducing a shear correction to both sides of
the equation would have been inexpedient. The single cantilever
deformation method would be applicable to the material, but the
dual cantilever sample holder allows for a more robust clamping
of the material. Since the two methods are principally identical,
only the dual cantilever system, see Fig. 2 was taken into consider-
ation in the scope of this work. All tests conducted within this
study were dynamic temperature sweeps applying an oscillating
force at multiple frequencies, carried out with the dual-cantilever
deformation system.
Margossian [5] reported to have encountered difficulties with
the dual-cantilever sample holder above and around melting tem-
perature, as the fixture appeared to be instable. Within the scope of
this work no instability issues were encountered with the sample
holder system while conducting the experiments.
The specimens were clamped in the fixtures using 30 cN m of
torque to tighten the bolts. The bolts were not retightened during
the experiments. The environmental chamber was heated to the
starting temperature with a heating rate of 10 K min1. To ensure
a homogeneous sample temperature, a 20 min balancing time
Fig. 2. Dynamic mechanical analysis: dual cantilever fixture.
 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
was used. The heating rate was set to 1.5 K min1 during the exper-
iments. The environmental chamber was continuously flushed
with nitrogen in order to avoid polymer oxidation at elevated
temperatures.
Temperature-sweeps were conducted from 150 °C to 260 °C.
150 °C was chosen as the lower limit of the investigated tempera-
ture range, as it is a favorable mold temperature for thermoform-
ing PA6 materials. It is very unlikely that the blank temperature
would drop below 150 °C during the swift forming process. The
upper limit was chosen to be 260 °C, because this is the upper limit
of the processing temperature range defined by the prepreg man-
ufacturer. At higher temperatures the matrix material is prone to
temperature degradation; furthermore, the mechanical properties
are not changing significantly once complete melting has occurred.
Six frequencies from 0.1 Hz to 10 Hz were investigated: 0.1, 0.5,
1, 2, 5 and 10 Hz. The maximum amplitude of the employed DMA
instrument is 0.24 mm. It takes a quarter of the cycle time for the
stamp to get from the initial position to the maximum amplitude.
At a 10 Hz frequency this means that the sample is flexed 0.24 mm
in 0.025 s, which roughly corresponds to a punch velocity of
10 mm s1. The actual strain rate of the blank during draping is
geometry dependent and non-uniform, but this has not been
accounted for in this study. This is a necessary simplification in
order to limit the complexity of the model.
Five experiments were made for both directions and both mate-
rial types. According to the material data sheet, there is a small dif-
ference between mechanical properties in weft and warp
directions. This can be attributed to the prepreg manufacturing
technique that pulls the weft rovings out of alignment by slightly
bending them. This difference was investigated by conducting
DMA measurements on samples cut in both directions. The result-
ing difference was deemed to be negligible, henceforth only warp
direction results were taken into account.
The result of DMA measurements are the complex modulus E
with its real and imaginary parts E0 and E00 which denote the stor-
age and loss modulus respectively.
E ¼ E0 þ ıE00
0
The storage modulus E describes the material’s ability to store
the deformation energy while the loss modulus E00 describes the
ratio of the deformation energy which is transformed into heat
by dissipation. Their ratio is usually referred to as tan d or damping.
E00
tan d ¼
E0
5.2. Rheometer bending experiments
The rheometer bending setup was used in the scope of this
work to investigate the bending behavior of single plies of Tepex
and Ultratape material. Additionally to the experimental procedure
performed by Sachs, relaxation was also investigated. After bend-
ing the specimen from 0° to 70° the bending angle was held at
70° for another 120 s, while measuring the reaction moment.
Fig. 3 gives an overview of the test procedure:
The tested parameters and their settings are listed in Table 2.
Three bending rates were tested at five temperatures. The rota-
Fig. 3. Rheometer bending test procedure.
25
Table 2
Investigated materials and parameters in rheometer bending tests.
Materials Temperatures (°C) Rotational velocities (rpm)
Tepex 60/150/210/220/ 0.1/1/10
Ultratape 210+/220+/260
tional velocities will further be referred to as 0.1 rpm (rotations
per minute), 1 rpm and 10 rpm.
Crystal melting is an endothermic transition, while crystalliza-
tion is exothermic. When cooling the environmental chamber from
above the melting temperature, the sample will consolidate and
crystallize, while giving off heat. This keeps the sample warm
and shifts the apparent melting temperature to a lower value.
Therefore, bending tests around melting temperature T m were per-
formed both while heating up and while cooling down after over-
heating. As expected, the experiments conducted while cooling the
sample from a temperature > T m delivered significantly lower
mechanical properties, as the material still exhibited molten state
behavior. Thus, tests were performed at 60 °C, 150 °C, 210 °C,
220 °C and 260 °C without overheating, furthermore, at 210 °C
and 220 °C with overheating to 255 °C and cooling back to 210 °C
and 220 °C (indicated with ‘‘+” henceforth and in Table 2), respec-
tively, to identify the shift in behavior due to crystallization and
crystal melting. After reaching the experimental temperature, a
two minute balancing time was employed to ensure a homoge-
neous temperature field in the sample during the test. The mea-
sured temperature, however, might differ from the actual sample
temperature during the measurements, due to the following
reasons:
 Temperature sensor was not inside/not in contact with the
sample.
 Heating rate was relatively high.
 Balancing time might not be sufficient to reach homogeneous
temperature field.
ð1Þ
6. Experimental results
In the following section the results of DMA bending and
rheometer experiments will be presented.
6.1. DMA experiments
ð2Þ
Twenty DMA measurements were made; five for both direc-
tions for both materials. As previously discussed, only warp direc-
tion results of the fabric and longitudinal results of the tape are
considered in the analysis. The storage modulus values of the five
identical measurements were averaged for every frequency tested.
Three frequencies are being further investigated: 0.1, 1 and 10 Hz.
These roughly correspond to the angular velocities used in the
rheometer measurements. Fig. 4 shows the average storage and
loss modulus as well as damping curves for these three frequen-
cies. The damping exhibits more distinctive frequency dependence
owing to the viscous behavior of the molten thermoplastic.
These averaged storage modulus values were then converted
into proportionality P i factors by dividing every value E0i with the
base value E0base , i.e. the storage modulus corresponding to the low-
est investigated temperature:
E0 ðT i Þ
PðT i Þ ¼ : ð3Þ
E0 ðT base Þ
Thus, an array of proportionality factors starts with a value of
one (corresponding to the base temperature) and gradually
decreases to a value less than one, which shows how much smaller
26 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
Fig. 4. Resulting storage modulus, loss modulus and damping for Tepex material at
0.1, 1 and 10 Hz.
the storage modulus is at the highest temperature investigated.
Such a set of proportionality factors describes the storage modulus’
temperature dependency at a given frequency. Multiplying the
proportionality factors with the bending stiffness measured at
the base temperature delivers a set of predicted bending stiffnesses
for the entire temperature range. These predicted values were
compared with the bending stiffness values calculated from the
TPRC rheometer measurement results, to validate the proposed
characterization method. Fig. 5 plots the six bending stiffness val-
ues derived from the rheometer bending results, along with the
predicted values delivered by the proposed DMA-based method.
Other studies [12,22,23] employing dynamic mechanical analy-
sis for material characterization found that the temperature and
frequency dependency of viscoelastic material properties can be
adequately studied with DMA. However, DMA derived storage
moduli are known to show considerable discrepancies to mechan-
ical testing moduli regarding their absolute values. Hobbibrunken
[23] calibrated the DMA derived moduli curves by shifting them
to a lower temperature to fit the mechanical testing results. Deng
[22] compared the DMA storage modulus to the Young’s modulus
measured by mechanical testing at the same temperature and
shifted all storage modulus values with the difference, producing
a shifted moduli curve. This latter shift method was applied to
the storage modulus of the Tepex material to demonstrate its inap-
plicability to thermoplastic materials investigated over a broad
temperature range involving transitions, as shown by the blue line
in Fig. 5(a). The proposed proportionality method overcomes these
issues for both material types, as proven by the comparison plotted
in Fig. 5. However, the validity of such manipulation, as also
Fig. 5. Comparison of predicted and experimental bending stiffness.
pointed out by Deng [22], needs to be assessed thoroughly involv-
ing a wide variety of materials and deformation methods, which
will be the scope of future studies.
The rheometer bending results yield the bending stiffness at
five investigated temperatures. Two temperatures, i.e. 210 °C and
220 °C were investigated both while heating up and while cooling
down from a temperature > T m , because these are located in the
transitional region of melting. Due to reasons discussed in Sec-
tion 5.2, discrepancies between the actual sample temperature
and the measured temperature might occur. Furthermore, the
melting temperature differs from the consolidation temperature,
as the results obtained at 210 °C and 220 °C with overheating, have
shown. Therefore, the bending stiffness values acquired at 210 °C
and 220 °C with overheating were shifted to 20 K higher tempera-
tures, to account for this discrepancy. The triangles in Fig. 5 repre-
sent the original bending stiffness values calculated from the
rheometer bending results, while the squares show their positions
after the shift.
This shift by 20 K of the overheated measurement points is nec-
essary due to the aforementioned uncertainties regarding the exact
specimen temperature as well as the definite difference regarding
the state of the thermoplastic matrix depending on whether the
samples were preheated to 255 °C or not. Plotting the DMA storage
modulus curve at both heating up and cooling down shows an
approximate 20 K discrepancy between the melting and the crys-
tallization temperature, as shown in Fig. 6. The same effect can
be observed on differential scanning calorimetry diagrams as the
offset between the endo-and exothermic peaks of melting and
crystallization, respectively. The effect is due to thermal inertia,
since the endothermic physical reaction of crystal melting is taking
up extra thermal energy while the exothermic crystallization is
heating the material from the inside [3]. This discrepancy was
determined separately for both Tepex and Ultratape materials
based on the DMA storage modulus curves, as seen in Fig. 6, and
was found to be approx. 20 K for both materials. This is the
expected result since both materials consist of the same PA6
matrix and the bending behavior was presumed to be matrix dom-
inated. Furthermore, the results in the transitional region are
highly dependent on factors such as heating rate and balancing
time. Therefore, the results do not correlate perfectly with the pre-
dicted curve, although exhibiting the same trend regarding tem-
perature dependency, see Fig. 5. The accurate measurement of
mechanical properties in this region requires further investigation.
Nevertheless, the molten state stiffness at 260 °C was successfully
predicted for both materials with outstanding accuracy.
Fig. 5 reveals that the bending stiffness curve derived from the
DMA storage moduli shows good correlation with the data
acquired from the rheometer bending measurements, especially
 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
Fig. 6. Difference between melting temperature and recrystallization temperature
of Tepex.
Fig. 7. Proportionality factors for both materials, Tepex and Ultratape, versus
temperature.
regarding the prediction of the molten state stiffness. Plotting the
proportionality factors for both the Tepex und Ultratape materials,
see Fig. 7, also proves that the change of the bending behavior over
temperature is matrix dominated. Thus, the proportionality factors
exhibit a nearly identical course in the investigated temperature
range, even though the bending stiffness of the Ultratape material
is an order of magnitude higher than that of the Tepex material.
Nonetheless, proportionality factors for bending stiffness predic-
tion were calculated separately for Ultratape and Tepex.
Further research is due on a wider range of materials to prop-
erly validate the method, but this first validation attempt proves
the method to be very promising. The proposed method has the
potential to significantly decrease the time and effort required
for the characterization of temperature-dependent mechanical
properties. Therefore, it will speed up the virtual prototyping pro-
cess while delivering higher quality simulation input data. It is
important to note that the method was applied to linear-elastic
bending stiffnesses. PAM-FORM among other commercial tools
uses an elastic approach for the bending stiffness, thus this method
is very valuable for such simulation tools.
6.2. Rheometer bending experiments
The rheometer exhibited a perfect velocity control even at the
highest investigated speed. The torque response, however, shows
a strong fluctuation in the initial region of the measurements, as
27
depicted in Fig. 8. This is due to the moment of inertia of the instru-
ment and the sample holder.
Time, bending angle and bending moment were documented
during the measurements. Three measurements were carried out
for every setting, the results of these have been averaged. The over-
all reproducibility was excellent, due to the outstanding velocity
control of the rheometer. The experimental results have proven
the materials’ strong temperature dependency, as exemplary
shown for the UD Ultratape material in Fig. 9 (a) and (b), displaying
the bending behavior below and above melting temperature,
respectively. The experiments also revealed the significant rate
dependency of the bending behavior for both investigated materials
above melting temperature. The material response is elastic at 60 °C
and 150 °C, i.e., beneath T m , and is therefore barely rate dependent,
as illustrated by Fig. 9 (c). The response at 210 °C and 220 °C with-
out overheating was also mostly elastic but has been disturbed by
the buckling of the material instead of a homogeneously distributed
bending. At 210 °C and 220 °C with overheating and 260 °C the
response showed a typical viscoelastic material behavior exhibiting
strong rate dependency and significant relaxation after deforma-
tion onto the same level of bending moment independent of the
velocity, see exemplary Fig. 9 (d). The fabric Tepex material showed
very similar behavior over the whole tested temperature and rota-
tional velocity ranges. Fig. 10 is an exemplary output for varying
temperature and 10 rpm velocity.
The bending angles have been used to calculate the correspond-
ing curvature j of the sample by:
a
j¼ ; ð4Þ
lb
where a is the bending angle in radiant and lb is the bending length
in mm. The initial bending length changes with the bending angle,
due to geometrical relations. The bending length was calculated
from the bending angle:

a
lb ¼ aR tan1 ; ð5Þ
2
where a is again the bending angle in radiant and R is the bending
radius in mm. The initial bending length (a ¼ 0) was 15 mm, since
the rheometer’s bending radius is 7.5 mm.
The bending stiffness B was calculated from the bending
moment M, curvature j and the area moment of inertia I:
M
B¼ : ð6Þ
jI
The calculated bending stiffness of both materials shows an ini-
tial unsettled section until a curvature of approximately 0.02–
0.03 mm1. This phenomenon is present irrespective of tempera-
ture and bending velocity. After this initial unsteady region, the
bending stiffness is linearly increasing below melting temperature
and slightly decreasing above T m . At 220 °C without overheating
the material experienced buckling, hence the distinct shape of
the curve. Therefore, approximately the first half of the bending
movement was neglected in the analysis. The bending stiffness val-
ues were thus averaged over the second half of the bending range
between 0.04 and 0.08 mm1 curvatures (a  35 . . . 60 ). This
range describes the long-term bending behavior.
Both DMA and rheometer experiments revealed highly vis-
coelastic material behavior in the molten state. The rheometer
bending results delivered bending moments which largely
depended on the deformation speeds. Higher deformation speeds
resulted in up to doubled bending moments with respect to lower
deformation speeds. Liang et al. [9] deemed the influence of the
bending stiffness on the wrinkling behavior in forming simulation
results as small, as only variations of one magnitude resulted in
visible differences in the used research-state draping simulation
tool. However, in a quantitative sensitivity study of the simulation
28 S. Ropers et al. / Composites: Part A 90 (2016) 22–32

Fig. 8. Fluctuations in moment signals due to inertia of instrument and sample holder.

Fig. 9. Influence of temperature and rotational velocity on bending behavior of the

Ultratape material. Fig. 10. Influence of temperature on bending behavior of the Tepex material.
 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
(a) Bending stiffness (b) Bending stiffness
base value multiplied by 2
Fig. 11. Sensitivity analysis on the bending stiffness: bending stiffness at base value
(a) and bending stiffness multiplied by 2 (b).
tool PAM-FORM [6], the influence of the bending stiffness even on
the shear angle of the examined fabric material was in the magni-
tude of the shear modulus. In order to estimate the influence of the
bending stiffness on the forming result, two simulations were run
in PAM-FORM on a half-dome geometry with only a difference in
bending stiffness of a factor of 2. The results are depicted in
Fig. 11. Doubling the bending stiffness results in a clearly different
forming result.
Therefore, the bending stiffness is one parameter which needs
to be thoroughly characterized in order to realistically predict
shearing and wrinkling of thermoplastic prepregs.
Thus, in order to further improve the predictability of draping
simulations, viscoelastic material models need to be implemented,
validated and fitted to corresponding experimental results. Sachs
[18] developed a viscoelastic bending model for thermoplastic
UD tapes based on their composition of thermoplastic matrix and
fiber layers. Yet the parameters had to be adapted to different con-
ditions as temperature or rotational velocity. Thus, in the following
section an attempt is made to find a general model for the material
behavior of thermoplastic prepregs under bending.
7. Viscoelastic material models for bending behavior
7.1. Linear viscoelastic model using Prony series
The DMA experiments reveal a clearly viscoelastic bending
behavior of thermoplastic prepregs. Polymers show a high depen-
dency of their mechanical properties on time/frequency and tem-
Fig. 12. Time–temperature superposition of the storage modulus.
29
perature. Ferry [25] described that an increase in temperature is
equal to a contraction of relaxation times with respect to the mate-
rial response to a load. The principle of time–temperature superpo-
sition (TTS) is based on this observation. As DMA machines are
usually equipped with a thermal chamber (cooled with nitrogen),
testing over a wide temperature range is possible while the neces-
sary time scale cannot be realized or would be inefficient. The
resulting dynamic moduli for each of the temperature points along
the temperature range are shifted by shift factors aT along the time
axis so that a continuous curve results, see following Fig. 12. This
curve is called master curve and defines the viscoelastic material
behavior at an arbitrary reference temperature. Corresponding
curves for the other temperature points are retrieved by a horizon-
tal shift of the master curve by the given temperature’s aT .
The shift factors are usually given by the Williams-Landel-Ferry
(WLF) equation [25]
C 1 ðT  T 0 Þ
log aT ¼  ; ð7Þ
C 2 þ ðT  T 0 Þ
or the Arrhenius equation. T 0 represents the chosen reference tem-
perature and C 1 and C 2 are material specific parameters to fit the
equation to the determined shift factors. The viscoelastic behavior
given by the master curve can be represented by a series of Prony
terms, which represent a series of Maxwell elements and a spring:
X
n
EðtÞ ¼ E1 þ Ei eðt=si Þ : ð8Þ
i¼1
E1 defines the long-term behavior, while each Ei has a corre-
sponding relaxation time si .
The principle of time–temperature superposition and the sub-
sequent description of the material behavior with Prony terms
and a shift function is usually applied for the transition from glassy
to rubbery behavior at the glass transition temperature T g for non-
continuous fiber reinforced polymers. As the measured continuous
fiber-reinforced polymers are still almost entirely elastic above T g
and the shape of the transition of the storage modulus at T g and T m
is similar, it was assumed that time–temperature superposition is
applicable for the transition to the molten state as well. Thus, for
the performed DMA measurements a Prony series with 20 ele-
ments (n ¼ 20) and a WLF representation of the shift factors were
derived for the Tepex fabric and the Ultratape UD material. These
were used to define a temperature-dependent linear viscoelastic
material model in Abaqus. The WLF function has a change of sign
at the temperature T sign ¼ T 0  C 2 . Abaqus uses for any tempera-
ture lower than T sign the same shift factor aT as for T sign .
30 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
Fig. 13. Simulated and experimental maximum bending moments at speed 10 rpm versus temperature.
The material model derived from the DMA measurements was
used to simulate the bending experiments performed with the
rheometer. As the material model is linear viscoelastic but the
experimental curves are highly non-linear only the maximum
bending moments at 70° deflection were compared. The simula-
tions were run at constant temperatures of 150 °C, 200 °C,
210 °C, 220 °C, 240 °C and 260 °C. The resulting maximum bending
moments, along with the bending moments derived from the
rheometer measurements are depicted in Fig. 13. The bending
moments obtained from the rheomter bending measurements
done at 210 °C and 220 °C with overheating were shifted to 20 K
higher temperatures to 230 °C and 240 °C, (similarly to those
shown in Fig. 5) for the same reasons discussed in Section 6.1.
For the Ultratape material the correlation between the simula-
tion results and the experimental moment values is excellent over
the whole investigated temperature range, especially at 210 °C and
220 °C, as shown in Fig. 13(b). (Notice that the y-axis of the dia-
grams have a logarithmic scale). The largest difference between
the results can be seen at 150 °C, since the WLF model was fit for
the range of 190 °C to 260 °C.
For the Tepex material all predicted maximum bending
moments are lower than the experimental results, although the
temperature dependency shows a very similar trend. This could
be due to the fact that the Prony series are derived from minuscule
strains from the DMA tests while the material exhibits a highly
non-linear bending behavior at high strains. The woven reinforce-
ment structure of Tepex results in a more complex bending behav-
ior due to the inherent undulation of the fiber rovings, while the
simple structure of the unidirectional reinforcement allows for a
behavior that is easier to predict. Nevertheless, the onset and rate
of stiffening with decreasing temperature is quite similar in exper-
iments and simulation. This comparison again justifies the trans-
ferability of the temperature dependency of the elastic moduli
derived from the DMA onto the bending stiffness.
7.2. Non-linear viscoelastic model using parallel rheological
framework
The parallel rheological framework (PRF) [26] in Abaqus enables
the definition of a customized viscoelastic material. Therefore, the
user can define any number of parallel viscoelastic elements to
model polymers that exhibit non-transient deformations and non-
linear viscous behavior under large deformations. Fig. 14 exhibits
an exemplary framework with four elements, which was used in this
work. For every element the share of the overall material response is
P
defined by a stiffness ratio S. The sum of the stiffness ratios S is
P
equal to or smaller than 1. If S equals unity no pure elastic element
P
”0” is added. In case the sum of the ratios S is smaller than 1, an
P
elastic element ”0” with the stiffness ratio 1  S is added.
In the scope of this work, for the viscous elements a Power-law
strain hardening model was used [27]:
Fig. 14. Schematic of parallel rheological framework with one hyperelastic element
”0” and three viscoelastic elements 1–3.
 n  1
_ cr ¼ Aq
~ ½ðm þ 1Þcr m mþ1 ; ð9Þ
where _ cr is the creep strain rate cr is the creep strain, q
~ is the devi-
atoric Kirchhoff stress A; m and n are parameters unique to the
material. The elastic elements are all specified by hyperelastic
material models which are isotropic and nonlinear. The strain
energy stored in the material per reference volume, the so-called
strain energy potential, describes the behavior of hyperelastic mate-
rials as a function of the strain.
For the description of the material behavior under bending of
the thermoplastic prepregs, a reduced polynomial form of the
strain energy potential was used, which has the general form
Xn
 i Xn
1
el 2i
U¼ C i0 I1  3 þ J 1 ; ð10Þ
i¼1 i¼1
Di
where n was set to 3 which then corresponds to the Yeoh model
[28]. U is the strain energy per unit of reference volume, C i0 and
Di are material parameters and I1 is the first deviatoric strain invari-
ant. For further details the reader is referred to the Abaqus manual
[29].
The material parameters of both the hyperelastic and the vis-
cous elements were fitted to the results of the rheometer bending
experiments. Therefore, the experimental bending angles and
bending moments were translated into strains and stresses. As
for high deformations there is no analytical correlation between
curvatures or bending moments and strains or stresses, respec-
tively. A tool was developed which creates the correlation itera-
tively. An optimization algorithm was used to fit the model
parameters to the corresponding strain stress curves of the differ-
ent rotational speeds. Thus, each subsequent iteration lets the
material model and thus the simulation results approach the
experimental results. The rheometer bending experiments were
modeled with solid elements using two elements over the thick-
ness, as the parallel rheological framework is limited to solid ele-
ments. Friction was neglected.
First, a simulation for each investigated bending speed (0.1 rpm,
1 rpm, 10 rpm) was run with the initial material parameters. The
simulation outputs were analyzed and stress, strain, reaction
 S. Ropers et al. / Composites: Part A 90 (2016) 22–32 31

moment and bending angle results were extracted for every incre-
ment. Thus, the correlation between logarithmic strain and bend-
ing angle as well as stress and reaction moment was established,
separately for each bending velocity. On the basis of this correla-
tion, strains and the ratios of stresses to reaction moments were
linearly interpolated corresponding to the bending angles of the
rheometer measurements. The resulting ratios of stress to reaction
moment were multiplied with the reaction moments from the
rheometer measurements, thus delivering the stress values corre-
sponding to the strains. The resulting stress–strain data was used
to calibrate the material model employed in the simulations to cor-
respond to the experimental results. The calibration was done via
an automated iteration process. The obtained material model has
then been implemented in the subsequent array of simulations.
This process was repeated until the simulation results showed an
acceptable correlation to the experiments. Such a process is shown
in Fig. 15. The same material model parameters were used for all
Fig. 16. Experimental and simulated bending moment of Tepex at 220 °C and 1 rpm
rotational velocities. The corresponding result for the 1 rpm speed rotational velocity.
is depicted in Fig. 16.
Figs. 15 and 16 clarify, that a non-linear viscoelastic material
model is capable of representing the actual material bending
behavior of thermoplastic prepregs. The results delivered by the
employed parallel rheological framework not only fit the experi-
mental results during the deformation of the specimen but also
accurately represent the relaxation behavior. The developed
method of coupling an iterative process with a calibration algo-
rithm allows for an efficient and robust determination of material
model parameters.

7.3. Combination of time–temperature superposition and non-linear

viscoelastic model

The method for determining the parameters of the non-linear

viscoelastic parallel rheological framework mentioned in the previ-
ous section was also successfully applied to the rheometer bending
results at a different temperature point, 210 °C. As the principle of
time–temperature superposition was deemed to be applicable for
Fig. 17. Experimental and simulated bending moment of Tepex at 210 °C and
the change in behavior over the temperature for the linear vis- 10 rpm rotational velocity.
coelastic material with a Prony series it was applied here as well.
The parameters for the parallel rheological framework for 220 °C
were used and each viscoelastic element was given a temperature
time shift in the form of the WLF equation and the corresponding
parameters C 1 and C 2 . The reference temperature T 0 was corre-
spondingly set to 220 °C. Fig. 17 exhibits the results at the
10 rpm velocity for a separate fit of the parallel rheological frame-
work parameters for 210 °C and the results of the shifted 220 °C

Fig. 18. Experimental and simulated bending moment of Tepex at 210 °C and 1 rpm
rotational velocity.

parameters. Both results are in correlation with the experimental

results. The shifted material model was also applied to the 1 rpm
velocity, see Fig. 18. As a temperature time shift is capable of
Fig. 15. Depiction of iterative process, experimental and simulated bending representing the material behavior at a temperature different to
moment of Tepex at 220 °C and 10 rpm rotational velocity. the reference temperature at 1 rpm and 10 rpm velocity, it is
32 S. Ropers et al. / Composites: Part A 90 (2016) 22–32
concluded that the change in behavior over the temperature can be
modeled using the behavior at a reference temperature and appro-
priate shift factors for the viscoelastic elements.
8. Conclusion
The dynamic mechanical analysis performed on the Tepex and
Ultratape material revealed the expected viscoelastic and
temperature-dependent material behavior. Although the derived
linear viscoelastic material model based on Prony terms using
time–temperature superposition could not represent the actual
bending behavior, it was possible to transfer the temperature
dependency of the storage moduli onto the actual bending stiffness
if linear elasticity was assumed. The applied shift in temperature
depending on the preheating needs to be further investigated.
Rheometer based bending experiments not only confirmed the
temperature dependency but also explained that the bending
behavior of thermoplastic prepregs is non-linear viscoelastic over
a wide curvature range. Additionally, a viscoelastic characteristic
relaxation could be observed. Finally, a method was developed to
calibrate a non-linear viscoelastic material model of Abaqus
(PRF) which correlates very well with the bending experiments.
However, the PRF is currently limited to solid elements and thus
not yet transferable onto draping simulations which are typically
using shell elements.
Future work needs to be conducted with respect to the transfer-
ability onto other materials and deformation modes, e.g. shear.
Also, non-linear viscoelastic material models need to be imple-
mented in draping simulation tools to verify the influence of such
models onto the draping simulation results.
References
[1] Mazumdar SK. Composites manufacturing: materials, product, and process
engineering. Boca Raton, Fla.: CRC Press; 2002. ISBN: 162870926X.
[2] Wang P, Hamila N, Boisse P. Thermoforming simulation of multilayer
composites with continuous fibres and thermoplastic matrix. Compos Part B:
Eng 2013;52:127–36.
[3] Guzman-Maldonado E, Hamila N, Naouar N, Moulin G, Boisse P. Simulation of
thermoplastic prepreg thermoforming based on a visco-hyperelastic model
and a thermal homogenization. Mater Des 2016;93:431–42.
[4] Boisse P, Hamila N, Vidal-Sallé E, Dumont F. Simulation of wrinkling during
textile composite reinforcement forming. Influence of tensile, in-plane shear
and bending stiffnesses. Compos Sci Technol 2011;71(5):683–92.
[5] Margossian A, Bel S, Hinterhoelzl R. Bending characterisation of a molten
unidirectional carbon fibre reinforced thermoplastic composite using a
dynamic mechanical analysis system. Compos Part A: Appl Sci Manuf
2015;77:154–63.
[6] Mallach A, Ropers S, Häusler F, Kardos M, Osswald TA, Gude M. Material
characterization and draping simulation of thermoplastic prepregs: the
influence of temperature. In: Society of plastics engineers – automotive
composites conference & exhibition; 2015.
[7] Haanappel SP. Forming of UD fibre reinforced
thermoplastics. Enschede: University of Twente; 2013. ISBN: 9036535018.
[8] Guzman-Maldonado E, Hamila N, Boisse P, Bikard J. Thermomechanical
analysis, modelling and simulation of the forming of pre-impregnated
thermoplastics composites. Compos Part A: Appl Sci Manuf 2015;78:211–22.
[9] Liang B, Hamila N, Peillon M, Boisse P. Analysis of thermoplastic prepreg
bending stiffness during manufacturing and of its influence on wrinkling
simulations. Compos Part A: Appl Sci Manuf 2014;67:111–22.
[10] Ropers S, Mabry P, Osswald TA, Mallach A. Draping simulation of
thermoplastic prepregs with special focus on the non-linear bending
stiffness. In: SPE ANTEC proceedings; 2015.
[11] Volkswagen Group Research. Unpublished research.
[12] Melo JDD, Radford DW. Time and temperature dependence of the viscoelastic
properties of CFRP by dynamic mechanical analysis. Compos Struct 2005;70
(2):240–53.
[13] Belendez Vazquez T, Neipp Lopez C, Belendez Vazquez A. Numerical and
experimental analysis of a cantilever beam: a laboratory project to introduce
geometric nonlinearity in mechanics of materials. Int J Eng Educ 2003;19
(6):885–92.
[14] Yu WR, Zampaloni M, Pourboghrat F, Chung K, Kang TJ. Analysis of flexible
bending behavior of woven preform using non-orthogonal constitutive
equation. Compos Part A: Appl Sci Manuf 2005;36(6):839–50.
[15] de Bilbao E. Bending properties of reinforcements in composites. In:
Composite reinforcements for optimum performance. Elsevier; 2011. p.
367–96. ISBN: 9781845699659.
[16] Martin TA, Bhattacharyya D, Collins IF. Bending of fibre-reinforced
thermoplastic sheets. Compos Manuf 1995;6(3–4):177–87.
[17] Peirce FT. 26—The handle of cloth as a measurable quantitiy. J Textile Inst
Trans 1930;21(9):T377–416.
[18] Sachs U. Friction and bending in thermoplasctic composites forming
processes. Enschede: University of Twente; 2014. ISBN: 946259483X.
[19] Menard KP. Dynamic mechanical analysis: a practical introduction. Boca
Raton, FL: CRC Press; 2008. ISBN: 978-1420053128.
[20] TA Instruments. TA instruments dynamic mechanical analyzer; 2002.
[21] NETZSCH-Geraetebau GmbH. Dynamic mechanical analysis: method,
technique, applications; 2013.
[22] Deng S, Hou M, Ye L. Temperature-dependent elastic moduli of epoxies
measured by DMA and their correlations to mechanical testing data. Polym
Testing 2007;26(6):803–13.
[23] Hobbiebrunken T, Fiedler B, Hojo M, Ochiai S, Schulte K. Microscopic yielding
of CF/epoxy composites and the effect on the formation of thermal residual
stresses. Compos Sci Technol 2005;65(10):1626–35.
[24] Bond-Laminates GmbH. Material data sheet: Tepex dynalite 102-RG600(x)/
47% roving glass – PA 6 consolidated composite laminate. Am Patbergschen
Dorn 11, 59929 Brilon, Germany; 2014, 04.12.2014.
[25] Ferry JD. Viscoelastic properties of polymers. 3rd ed. New York: Wiley; 1980.
ISBN: 9780471048947.
[26] Lapczyk I, Hurtado JA, Govindarajan SM. A parallel rheological framework for
modeling elastomers and polymers. American chemical society – 182nd
technical meeting rubber division; 2012.
[27] Nandi B, Dalrymple T, Yao J, Lapczyk I. Importance of capturing non-linear
viscoelastic material behavior in tire rolling simulations. The tire society –
annual meeting on tire science and technology; 2014.
[28] Martins PALS, Natal Jorge RM, Ferreira AJM. A comparative study of several
material models for prediction of hyperelastic properties: application to
silicone–rubber and soft tissues. Strain 2006;42(3):135–47.
[29] Dassault Systemes SIMULIA Corp., Providence, RI, USA. Abaqus 6.14 Manual;
2014.